Encoding Discordance in the Alzheimer's Disease A/T/N Framework

Area of Science:

• Neuroscience
• Biomarker Research
• Artificial Intelligence in Medicine

Background:

• The amyloid/tau/neurodegeneration (A/T/N) framework is widely used in Alzheimer's disease (AD) research for staging.
• Existing methods using the A/T/N framework often compromise between adaptivity and interpretability.
• There is a need for advanced methods to improve early AD detection and prediction.

Purpose of the Study:

• To develop an interpretable, hybrid method for predicting incident Alzheimer's disease (AD).
• To leverage the A/T/N framework in a novel, data-driven yet interpretable manner.
• To enhance the accuracy and early detection capabilities for AD.

Main Methods:

• Introduced Neurosymodal Data Fusion, an interpretable, hybrid approach.
• Encoded the A/T/N framework as a logic program.
• Utilized neural networks to extract input biomarker features for AD prediction.

Main Results:

• Achieved up to 0.84 sensitivity in predicting four-year incident AD.
• Generated scores for A/T/N profiles, indicating relative importance for predictions.
• Identified potential limitations of current empirical cut-off values for A and T biomarkers in the ADNI dataset.

Conclusions:

• The developed pipeline offers a novel application of the A/T/N framework.
• This method has the potential to significantly improve early AD screening.
• The approach could enable earlier identification of AD years before clinical symptoms manifest.